Clinical Trials - analysis End-points, planned and unplanned secondary analyses, interim analyses, stopping rules Dick Menzies, MD Respiratory Epidemiology and Clinical Research Unit Montreal Chest Institute TB Research methods course July 9, 2015

Interim analyses Final - Descriptive analysis Participation – the Consort diagram (Figure 1) Study participants (Table 1) Primary analysis (superiority vs non-inferiority) Effectiveness (Intention to treat ) Modified intention to treat Efficacy – per protocol Secondary analyses Planned and Hypothesis generating Lecture 4: Data analysis Overview

Interim Analysis and Stopping Rules In large trials interim analyses commonly done. Adverse events – Primary outcomes - Can the study be ended early – hypothesis answered. Or, Should the study be ended early – patient’s safety. Must use more stringent rules (p<0.01, not p<0.05) Usually reviewed by independent panel (DSMB)

Interim Analysis Example Vernon et al Lancet 1999 Enrolment began in April By early 1997 four HIV positive patients had relapsed with Rifampin mono resistance. All occurred in those taking once weekly RPT-INH. The DSMB, CDC, and the investigators decided to stop enrolment of HIV positive patients. Those still taking once weekly RPT-INH were switched to standard treatment. Once weekly INH-RPT Twice weekly INH-RIF p value Number Relapse RIF-R 40.05

Final Analysis: Step 1 – Accounting for all subjects The CONSORT statement - JAMA 1996 (consolidated standards of reporting trials) Revised CONSORT Statement. Ann Intern Med 2001; vol 134: p666

Consort diagram example: Oflutub trial Potential eligible Population Not randomized Randomized ITT population MITT populations Valid exclusions post-randomization Did not complete Tx Did not complete F-U Per protocol Population

Step 1B: Analysis of non- participants Subjects who are screened as potential participants, but were not eligible, or refused. If not randomized do not impact the internal validity of the study. But affect external validity (capacity to generalize). Especially important if high exclusion or refusal rate

Step 2: Describing and comparing study participants (Table 1) This is a simple descriptive analysis comparing study participants randomized to the different interventions Demographic characteristics (age and sex) Major clinical characteristics (extent of disease, drug resistance) Comorbidities (HIV, Diabetes etc) No statistical testing please

Baseline characteristics – example Swaminathan 2010 varying lengths of treatment in HIV TB Characteristic of Study SubjectsReg6M (n=167)Reg9M (n=160) Median age, years (IQR)33 ( (29-39) Median weight, kg (IQR)44(39-50) Median CD4 cells/mm (IQR)152 (80-304)167 (88-280) Median viral Load, (copies/ml)94,300 (n=100)168,000 (n-113) Males N % 119 (79%)112 (75%) Pulmonary TB (n=299) Culture Positive117 (78%)110 (74%) Susceptible to all first-line drugs,99 (88%)95 (88%) Culture Negative34 (22%)38 (26%) Extrapulmonary TB (n=28) Culture Positive4 (25%)2 (16%) Culture Negative12 (75%)10 (84%)

Baseline characteristics – example Oflutub trial Did the randomization work? Patient Characteristics Regimen Gatifloxacin n=136Moxifloxacin n=115Control n=165 Sex Male103 (76%)83 (72%)122 (74%) Age (years): <4090 (66%)88 (77%)120 (73%) Body weight (Kg): Mean Sputum culture 3+ growth107 (79%)94 (82%)127 (77%) X-ray Chest >2 Zones affected107 (79%)94 (82%)127 (77%)

Step 3: Primary analysis The primary analysis addresses the primary objective. Sample size calculations were based on this planned analysis.

Primary analysis All randomized participants included. Withdrawals : Participants who sign consent, and are randomized. But withdraw consent – so ethically not included in the analysis. Can bias the results of the study (the 2 groups of participants remaining may not be comparable) Drop-outs from therapy : Do not complete therapy, but do complete follow-up post therapy. Contribute fully to analysis. VERY IMPORTANT. Let people drop-out/stop therapy BUT do not lose them Lost – no idea of final outcome. More difficult

Superiority Studies (reminder) Test New Interventions against a standard or placebo. Hypothesis: New intervention is better. New intervention will be adopted if patients’ outcomes are better.

Superiority studies: Results: CAN conclude superiority Effect ofTrue Effect of Standard TherapyNew therapy 0 No Effect

Superiority studies: Results: CANNOT conclude superiority 0Effect of True Effect of No EffectStandard New Therapy Therapy

Non-inferiority Studies If current therapy is efficacious - But is very costly, or lengthy (poor completion) - Or has major side effects Alternate therapy is cheaper, or shorter, or safer. Then we want to show that the efficacy of new treatment is not worse. This is called a Non-inferiority study.

Non-Inferiority studies - Results CAN conclude non-inferiority No effectLeast Acceptable Effect of Effect of NewStandard Therapy Therapy

Non-Inferiority studies - Results CANNOT conclude non-inferiority No effectLeast Acceptable Efficacy of Efficacy of NewStandard Therapy Therapy

Efficacy vs Effectiveness Efficacy (per protocol) : The extent to which a specific intervention, procedure, regimen, or service produces a beneficial result under ideal conditions; This means the patient actually took all doses of treatment, And all elements of the protocol followed (ie full follow-up) Optimal Estimate: Answers the patient’s question “Doctor, if I take this drug, will I get better?”

Efficacy vs Effectiveness Effectiveness (intention to treat) The effect of a specific intervention, procedure, regimen, or service, when deployed in the field in routine circumstances. This accounts for non-compliance, dropouts and side effects. All patients randomized (allocated to treatment) are analysed, whether or not they completed the prescribed regimen, and follow-up. Conservative estimate: Answers the public health question “What is the overall effect of this treatment given to a population?”

Duration of INH Therapy and efficacy/effectiveness (IUAT trial - Patients with Fibrotic Lesions) PopulationDurationReduction in TB All participants INH 12 mo.75% (Effectiveness)INH 6 mo.65% INH 3 mo.21% Completer/compliers INH 12 mo.93% (Efficacy)INH 6 mo.69% INH 3 mo.31% Bull WHO 1982;555-64

ITT and MITT Analyses: example Sterling et al; 3HP vs INH; NEJM 2011 Study GroupNSubjects with Active TB Difference in Cumulative Rate no. no. per patient yr Cumulative rate percentage points Modified intention-to-treat analysis 9 mos INH INH-RPT Per-protocol analysis 9 mos INH INH-RPT

Mis-use of ITT analysis The goal of ITT analysis is to produce realistic estimates of what the treatment will achieve in real life. BUT, many RCT select subjects carefully on the basis of compliance - Baseline characteristics (lifestyle, employment, etc) - Run-in period – often 1-3 months to assess compliance What effect does this have on ITT analysis?

Modified Intention to treat Analysis (MITT) There may be instances where patients may need to be randomized before all information is available. Protocol may specify valid exclusions post randomization. Particularly common in TB trials when eligibility depends upon culture and/or drug susceptibility testing – since TB treatment cannot wait. In latent TB trials, household contacts may start LTBI therapy before knowing the DST of the index cases – again because treatment initiation should not wait

Secondary Analyses: Planned Many studies pre-specify planned secondary analysis This should be stated in the published study protocol For this can pre-specify that some subjects will be excluded To examine effect of age or gender, or comorbidities Can examine different end-points – Adverse events Per protocol analysis is a commonly planned secondary analysis

Planned Primary and Secondary analyses – example Gler et al, Use of Delanamid for MDR-TB; NEJM, 2013

Planned Primary and Secondary analyses – example Gler et al Delamanid for MDR TB NEJM 2012 Primary endpoint – proportion with sputum culture conversion at 2 months – MITT Multiple secondary endpoints assessed. These included time to sputum culture conversion Safety performed in all patients randomized who received at least one dose of study medication (ITT) All endpoints pre-specified in formal statistical analysis plan. Plan finalized and filed before analysis begun.

Consort diagram 96 Excluded – not eligible (91) - other (5) 161 DMD 100 BID 161 in ITT (Safety) 20 Excluded Negative Culture Not MDR 141 in MITT (2 months culture conversion ) 160 DMD 200 BID 160 in ITT (Safety) 35 Excluded Negative Culture Not MDR 125 in MITT (2 months culture conversion) 611 patients assessed for Eligibility 34 Refused 481 Randomized 160 Placebo 160 in ITT (Safety) 14 Excluded 6 Withdrew 6 SAE 2 Lost 122 in Efficacy (time to culture conversion) 18 Excluded 14 withdrew 4 SAE 0 Lost 123 in efficacy (time to culture conversion) 15 Excluded 8 Withdrew 4 SAE 3 Lost 120 in efficacy (time to culture conversion ) 24 Excluded Negative Culture Not MDR 136 in MITT (2 months culture conversion)

Delamanid 100mg Twice Daily (N=161) Delamanid 200mg Twice Daily (N=161)Placebo (N=160) Anemia18(11.2)10(6.2)14(8.8) Nausea58(36.0)65(40.6)53(33.1) Prolonged QT interval on ECG16(9.9)21(13.1)6(3.8) Paresthesias17(10.6)20(12.5)12(7.5) Anorexia23(14.3)34(21.2)24(15.0) Hypokalemia20(12.4)31(19.4)24(15.0) Planned secondary analysis: Incidence of Adverse Events Uses ITT population (took at least 1 dose of study drug)

Primary analysis - Uses MITT population: 2 Month culture Conversion on MGIT Planned secondary Analysis - Uses MITT population: 2 mos conversion on solid media

Planned secondary Analysis – Efficacy: Uses per protocol Population Time to culture conversion

Planned Primary and Secondary analyses – example Use of TMC-207 (Bedaquiline) for MDR TB Diacon et al NEJM 2009

Bedaquiline for MDR TB Diacon et al NEJM 2009 Primary Analysis (MITT)

Adverse EventTMC207 (N=23)Placebo (N=24) Nausea6(26)1(4) Diarrhea3(13)1(4) Arthralgia4(17)3(12) Rash2(9)4(17) Dizziness3(13)2(8) Secondary Analysis (ITT) Incidence of Adverse Events

Planned Secondary Analyses: (Efficacy) Rate of bacterial killing (per protocol)

Secondary Analyses: Post hoc (Hypothesis generating) Hypothesis generating vs data dredging Once the primary and planned secondary analyses are done, Then many exploratory analyses can be performed Risks- If 20 tests are done, 1 will be significant at p<.05 by chance alone. Especially if not clearly driven by a priori hypotheses, but rather by a desire for a p<.05!! Advantages- RCT generate a wealth of data which can and should be used to address other questions -but very important to describe these analyses clearly as such.

Post hoc Analyses – example DMD Improves outcomes and reduces mortality in MDR TB Skripconoka et al, ERJ 2013 What they wrote in Abstract - Results and Conclusions: “Mortality was reduced to 1% on those receiving long-term DMD vs short-term or no DMD (8.3% p>.001)” “Treatment benefit was also seen among patients with XDR TB” “This analysis suggests that treatment with DMD for 6 months in combination with optimized background regimen can improve outcomes and reduce mortality in patients with both MDR and XTR TB”

Post hoc Analyses – example DMD Improves outcomes and reduces mortality in MDR TB Skripconoka et al, ERJ 2013 Methods: Follow-up study after conclusion of initial 2 month treatment study Study launched 2-12 months after end of first study Substantial intervals between initial 2 month treatment with DMD, and later treatment Patients not randomized. Less than half selected for DMD by provider or by themselves.

24 month outcomes after treatment with DMD plus OBR in patients with MDR or XDR Skripconoka et al, ERJ 2013 Treatment Outcome 6-8 months DMD N= Months DMD N=229 Cured110 (57%; 50-64)111 (48%; 42-55) Completed33 (17%; 12-23)15 (7%; 4-11) Died2 (1%; 0.1-4)19 (8%; 5-13) Failed32 (17%; 12-23)26 (11%; 8-16) Defaulted15 (8%; 4-13)58 (25%; 20-32)

24 month outcomes after treatment with DMD plus OBR in patients with XDR only. Skripconoka et al, ERJ 2013 Treatment Outcome 6-8 months DMD N= Months DMD N=12 Cured11 (25%; 13-40)5 (42%; 15-72) Completed16 (36%; 22-45)1 (8%; ) Died0 (0 )3 (24%;5-57) Failed14 (32%; 19-48)3 (25%; 5-57) Defaulted3 (7%; 1-19)0 (0)

Does the abstract reflect the design of the study? Does the abstract reflect the strength of the findings?

Thanks

Consort diagram example Mfinanga LID 2014